AI has become a sort of corporate mantra, and machine learning (ML) and gen AI have become additions to the bigger conversation. The role of CIO, especially, has had to adapt accordingly, as demonstrated by Euronics, the Amsterdam-based international electrical retail association. Here, the work of digital director Umberto Tesoro started from the need to better use digital data to create a heightened customer experience and increased sales.
Gartner suggests extending the data and analytics strategy to include AI and avoid fragmented initiatives without governance. Gen AI in practice is a special case of Euronics’ strategy that concerns data and analysis, and the task of those who direct it — the CIO or the CDO — is to understand when to apply it, and when not to.
In fact, gen AI isn’t currently among the implemented technologies at Euronics because Tesoro doesn’t see use cases functional to the retail activity. “IT must be at the service of the business,” he says.
The first step of the manager’s team was instead to hire a UX designer to not only design the interface and experience for the end user, but also carry out tests to bring qualitative and quantitative evidence on site and app performance to direct the business.
“E-commerce is a journey that goes from visiting the site to completing the purchase,” says Tesoro. “We monitor the entire flow and use aggregated data to evaluate the best solutions and experience to bring to the customer. We always present consumers with two different experiences and evaluate the result. From there, we choose one and make any changes to the site experience, so our strategy for e-commerce is entirely data-driven.”
The site solutions are strengthened by ML, which helps deliver personalized content by suggesting the display of related products in line with those already purchased by customers. For ML and analytics, Tesoro purchases products on the market it considers best in class, and the applications are then customized through technological partners for the needs of the company’s site and app.
“The internal IT team must be able to govern and guide the supplier, but we don’t do in-house development,” he says. “I don’t consider it convenient in our case. I give directions and strategies to the supplier and the partner, and an internal project manager acts as a link. For us, the key figures of the digital team are the UX designer and the business analyst because internally, we work on strategic objectives: customer experience and data analysis to support sales.”
It’s the valorization of the company’s data assets, and the goal, to constantly improve CX that led to the recent partnership between Euronics and commerce media company Criteo to implement Criteo’s retail media solutions on Euronics’ e-commerce site.
“We provide the consumer with a product that’s in line with a specific search on the site,” says Tesoro. “There are a variety of advantages for us: We give a better result to the advertiser and we create the conditions for a certain product to be sold on our e-commerce platform. We have a positive effect on sales thanks to the analysis of data on the consumer’s search intent provided by the Criteo platform.”
The value of data in nonprofits
Even for Emergency, the Italian NGO, data is a strategic asset to be enhanced and protected.
“Data is the support for the core activity of hospitals,” says CIO Manuele Macario. “It must always be safe for the people we treat.”
This philosophy has led to the activation of an information system that manages clinical data in the three Emergency surgical centers in Afghanistan through the SDC software platform. This has an architecture structured on open-source components, both on the servers and tablets distributed among various Emergency clinics.
“The open-source software platform was created by our medical division to have a record that could work even in precarious conditions,” says Macario. “For example, it adapts to the absence of an internet connection and offline work. The data is then re-transported when the line is available. It’s easy to install and we can transfer it to other sites.”
This doesn’t detract from the fact it’s a very advanced clinical data collection system since it’s digital, in real time, and secure because the data is encrypted on VPN and sent to Emergency’s central data center in Milan. It’s here where analysis through Microsoft Power BI is also performed. The infrastructure is on-prem, Macario explains, to preserve an investment made, but not to preclude the cloud. In fact, several steps have been taken in recent years to migrate some services to the cloud.
“The important thing in data management is having a solid disaster recovery plan,” says Macario. “In fact, security for an NGO like ours is both a cyber and physical problem because not only are we the target of attacks, but we operate in war zones, where the services provided aren’t always reliable and, in the event of failures, hardware replacement parts are difficult to find.”
Innovative encryption and geographic data backup technologies are applied, in particular immutable cloud technology that protects against ransomware. These are supported by AI for endpoint protection. User identities are also managed on the Azure Entra ID platform, which has integrated AI and warns of suspicious activity in real time.
The turning point in data management
But the real change in the use of AI for Emergency came with gen AI applied in the Amanat project, also carried out in Afghanistan. The initiative is based on what was done eight years ago with the scanning of over 10 million sheets of medical records.
The next step of the project was to use the digitalized data and analyze it. The scanned data, in fact, weren’t readable by software because they were extracted from documents from a hospital that operates in trauma and, therefore, often written in haste with inaccurate and contracted handwriting.
“We turned to the big technology players to solve the problem and the LLM algorithms led to a turning point, because they allowed us to carry out the analyses,” says Macario. “These are used by our Medical Division departments to analyze access to care and improve quality, obtain statistics, create an archive, and understand what instruments, drugs, and doctors we need in a war context. The data form a scientific basis on which to base our intervention and our ability to report the effects of war on civilian populations.”
Macario’s team then started a PoC by randomly taking a thousand medical records from 2002 to 2018 from the trauma hospital in Kabul. The PoC was necessary to test the technology’s functioning and contain costs, which, by applying the system to all documents, would have grown significantly.
“We chose Microsoft and Azure OpenAI technology as our partner,” he says. “We didn’t want to put the medical record data on the open OpenAI-ChatGPT system. Instead, we used space on our Microsoft tenant, which guaranteed us the privacy and protection of patient data.”
Microsoft’s technology was customized, providing computational power, but the Emergency team needed to customize the Azure Document Intelligence algorithm by training it on the formats used in medical records, and making the AI understand where and what information to interpret.
From there, a searchable relational database was created from which to perform a posteriori analysis of the activity to capture trends. To read the digitized medical records, Macario’s team created specific prompts inserted into Azure OpenAI to obtain, for example, the unpacking of acronyms or the interpretation of only partially written words.
“We isolated information about how much time passed between the injury and the hospital treatment, which is crucial to understand where to place our first-aid centers and whether our surgery is effective,” Macario explains. “With the right prompts, we guided Azure OpenAI to correct and convert this type of information, providing us with data that can be analyzed and visualized on graphs.”
The right amount of investment in gen AI
Macario emphasizes how gen AI should be applied only if there is a benefit that justifies the investment.
“I don’t think of using AI and LLMs other than for what our organization really needs, otherwise it’s an unsustainable technology in terms of costs and burden on the environment,” he says. “For me, AI and gen AI should be used to obtain a result that couldn’t be obtained otherwise, or because a significant benefit is expected. This is the case of our Amanat project, which used gen AI not only to provide useful data for our operations, but to be ready to best face a future intervention in a war context, and be able to tell the consequences of that war with data.”
Euronics’ Tesoro is also selective with gen AI as the company currently doesn’t apply it to retail activity, but is carrying out tests in the Salesforce ecosystem to understand possible uses of it for productivity. And here the prospects seem promising.
“I don’t see gen AI as having a big impact on our business at this stage, but I’m interested in automating manual processes, and we’re already testing productivity tools with generative AI for repetitive and non-valuable tasks,” says Tesoro. “For me, AI is a tool to empower people, allowing them to not waste their talent on mechanical tasks and instead be shifted to those that enhance their intellectual abilities.”
Macario reiterates that in Emergency’s Amanat project, ChatGPT’s task has been carefully circumscribed. “It’s a tool, not an oracle, and it must be given boundaries, a clear indication of what it must do,” he says. “The algorithms speak through statistics. They don’t give a right or wrong answer, but one that has a more or less high degree of reliability. Below a certain threshold, however, the answer is not acceptable. Of the thousand records put in the PoC, we discarded half as unreliable.”
Don’t forget traditional AI
According to author and data and analytics expert Stefano Gatti, gen AI helps but it’s not mature enough to manage customer-facing services. “Instead, it’s mature enough to support the increase in internal productivity, as several CIOs recognize,” he says. “Human supervision or, in any case, verification of the reliability of the result remain fundamental.”
Gartner writes in a similar sense that CIOs should first understand whether the use case creates value for the business and is feasible in practice before investing in gen AI, since it’s difficult to justify applying it indiscriminately. There are also established, or traditional, AI techniques that carry less risk such as optimization, simulation, and knowledge graphs that can be effective without gen AI.